# ODICE: Revealing the Mystery of Distribution Correction Estimation via Orthogonal-gradient Update

arXiv: https://arxiv.org/abs/2402.00348 · HTML: https://mlsanity.com/paper/2402.00348/

## Key facts

- Cited by 21 papers in the map (data through 14 Oct 2026).
- Placed in Offline RL algorithms, one of 13 directions in Offline RL and imitation learning from data.
- Builds on 55 of its 69 references that are in the map.
- First posted on arXiv 1 Feb 2024.

## Placed in

- Offline RL and imitation learning from data / [Offline RL algorithms](https://mlsanity.com/offlinerl/offline-rl-algorithms/)
- Deep reinforcement learning methods / [Deep RL algorithms and representations](https://mlsanity.com/deeprl/deep-rl-algorithms-and-representations/)

## Builds on (most cited first, 30 of 55)

- [Layer Normalization](https://mlsanity.com/paper/1607.06450/) (2016)
- [Playing Atari with Deep Reinforcement Learning](https://mlsanity.com/paper/1312.5602/) (2013)
- [Convex optimization](https://mlsanity.com/paper/2106.01946/) (2021)
- [Addressing Function Approximation Error in Actor-Critic Methods](https://mlsanity.com/paper/1802.09477/) (2018)
- [Improved Training of Wasserstein GANs](https://mlsanity.com/paper/1704.00028/) (2017)
- [Conservative Q-Learning for Offline Reinforcement Learning](https://mlsanity.com/paper/2006.04779/) (2020)
- [D4RL: Datasets for Deep Data-Driven Reinforcement Learning](https://mlsanity.com/paper/2004.07219/) (2020)
- [Offline Reinforcement Learning with Implicit Q-Learning](https://mlsanity.com/paper/2110.06169/) (2021)
- [Off-Policy Deep Reinforcement Learning without Exploration](https://mlsanity.com/paper/1812.02900/) (2018)
- [A Minimalist Approach to Offline Reinforcement Learning](https://mlsanity.com/paper/2106.06860/) (2021)
- [Stabilizing Off-Policy Q-Learning via Bootstrapping Error Reduction](https://mlsanity.com/paper/1906.00949/) (2019)
- [Generalized Out-of-Distribution Detection: A Survey](https://mlsanity.com/paper/2110.11334/) (2021)
- [AWAC: Accelerating Online Reinforcement Learning with Offline Datasets](https://mlsanity.com/paper/2006.09359/) (2020)
- [Reinforcement Learning with Deep Energy-Based Policies](https://mlsanity.com/paper/1702.08165/) (2017)
- [Behavior Regularized Offline Reinforcement Learning](https://mlsanity.com/paper/1911.11361/) (2019)
- [Uncertainty-Based Offline Reinforcement Learning with Diversified Q-Ensemble](https://mlsanity.com/paper/2110.01548/) (2021)
- [Critic Regularized Regression](https://mlsanity.com/paper/2006.15134/) (2020)
- [DualDICE: Behavior-Agnostic Estimation of Discounted Stationary Distribution Corrections](https://mlsanity.com/paper/1906.04733/) (2019)
- [Offline Reinforcement Learning with Fisher Divergence Critic Regularization](https://mlsanity.com/paper/2103.08050/) (2021)
- [MT-Opt: Continuous Multi-Task Robotic Reinforcement Learning at Scale](https://mlsanity.com/paper/2104.08212/) (2021)
- [IQ-Learn: Inverse soft-Q Learning for Imitation](https://mlsanity.com/paper/2106.12142/) (2021)
- [AlgaeDICE: Policy Gradient from Arbitrary Experience](https://mlsanity.com/paper/1912.02074/) (2019)
- [Imitation Learning via Off-Policy Distribution Matching](https://mlsanity.com/paper/1912.05032/) (2019)
- [Offline RL Without Off-Policy Evaluation](https://mlsanity.com/paper/2106.08909/) (2021)
- [Pessimistic Bootstrapping for Uncertainty-Driven Offline Reinforcement Learning](https://mlsanity.com/paper/2202.11566/) (2022)
- [GenDICE: Generalized Offline Estimation of Stationary Values](https://mlsanity.com/paper/2002.09072/) (2020)
- [OptiDICE: Offline Policy Optimization via Stationary Distribution Correction Estimation](https://mlsanity.com/paper/2106.10783/) (2021)
- [Extreme Q-Learning: MaxEnt RL without Entropy](https://mlsanity.com/paper/2301.02328/) (2023)
- [Offline RL with No OOD Actions: In-Sample Learning via Implicit Value Regularization](https://mlsanity.com/paper/2303.15810/) (2023)
- [Neural Photo Editing with Introspective Adversarial Networks](https://mlsanity.com/paper/1609.07093/) (2016)

## Built on by (most cited first, 21 of 21)

- [Diffusion-DICE: In-Sample Diffusion Guidance for Offline Reinforcement Learning](https://mlsanity.com/paper/2407.20109/) (2024)
- [Offline Reinforcement Learning with OOD State Correction and OOD Action Suppression](https://mlsanity.com/paper/2410.19400/) (2024)
- [Offline-Boosted Actor-Critic: Adaptively Blending Optimal Historical Behaviors in Deep Off-Policy RL](https://mlsanity.com/paper/2405.18520/) (2024)
- [A Dual Approach to Imitation Learning from Observations with Offline Datasets](https://mlsanity.com/paper/2406.08805/) (2024)
- [ComaDICE: Offline Cooperative Multi-Agent Reinforcement Learning with Stationary Distribution Shift Regularization](https://mlsanity.com/paper/2410.01954/) (2024)
- [Dichotomous Diffusion Policy Optimization](https://mlsanity.com/paper/2601.00898/) (2025)
- [Data Center Cooling System Optimization Using Offline Reinforcement Learning](https://mlsanity.com/paper/2501.15085/) (2025)
- [An Optimal Discriminator Weighted Imitation Perspective for Reinforcement Learning](https://mlsanity.com/paper/2504.13368/) (2025)
- [Markov Balance Satisfaction Improves Performance in Strictly Batch Offline Imitation Learning](https://mlsanity.com/paper/2408.09125/) (2024)
- [Semi-gradient DICE for Offline Constrained Reinforcement Learning](https://mlsanity.com/paper/2506.08644/) (2025)
- [Balance Equation-based Distributionally Robust Offline Imitation Learning](https://mlsanity.com/paper/2511.07942/) (2025)
- [SEMDICE: Off-policy State Entropy Maximization via Stationary Distribution Correction Estimation](https://mlsanity.com/paper/2512.10042/) (2025)
- [Dual Alignment Maximin Optimization for Offline Model-based RL](https://mlsanity.com/paper/2502.00850/) (2025)
- [Learning What to Do and What Not To Do: Offline Imitation from Expert and Undesirable Demonstrations](https://mlsanity.com/paper/2505.21182/) (2025)
- [Value from Observations: Towards Large-Scale Imitation Learning via Self-Improvement](https://mlsanity.com/paper/2507.06701/) (2025)
- [Symmetric Behavior Regularized Policy Optimization](https://mlsanity.com/paper/2508.04225/) (2025)
- [Mildly Conservative Regularized Evaluation for Offline Reinforcement Learning](https://mlsanity.com/paper/2508.05960/) (2025)
- [Efficient Anti-exploration via VQVAE and Fuzzy Clustering in Offline Reinforcement Learning](https://mlsanity.com/paper/2602.07889/) (2026)
- [[Re] FairDICE: A Fair Tradeoff in Multi-objective Offline RL](https://mlsanity.com/paper/2603.03454/) (2026)
- [When Dynamics Shift, Robust Task Inference Wins: Offline Imitation Learning with Behavior Foundation Models Revisited](https://mlsanity.com/paper/2605.17017/) (2026)
- [FedGuide: Diffusion Prior Alignment and Value Baseline Guidance for Heterogeneous Federated Reinforcement Learning](https://mlsanity.com/paper/2609.18964/) (2026)

## Abstract (arXiv)

In this study, we investigate the DIstribution Correction Estimation (DICE) methods, an important line of work in offline reinforcement learning (RL) and imitation learning (IL). DICE-based methods impose state-action-level behavior constraint, which is an ideal choice for offline learning. However, they typically perform much worse than current state-of-the-art (SOTA) methods that solely use action-level behavior constraint. After revisiting DICE-based methods, we find there exist two gradient terms when learning the value function using true-gradient update: forward gradient (taken on the current state) and backward gradient (taken on the next state). Using forward gradient bears a large similarity to many offline RL methods, and thus can be regarded as applying action-level constraint. However, directly adding the backward gradient may degenerate or cancel out its effect if these two gradients have conflicting directions. To resolve this issue, we propose a simple yet effective modification that projects the backward gradient onto the normal plane of the forward gradient, resulting in an orthogonal-gradient update, a new learning rule for DICE-based methods. We conduct thorough theoretical analyses and find that the projected backward gradient brings state-level behavior regularization, which reveals the mystery of DICE-based methods: the value learning objective does try to impose state-action-level constraint, but needs to be used in a corrected way. Through toy examples and extensive experiments on complex offline RL and IL tasks, we demonstrate that DICE-based methods using orthogonal-gradient updates (O-DICE) achieve SOTA performance and great robustness.